Overview

The dashboard home shows live operational state: request and token rates, a model-residency timeline, currently running models, upstream and GPU health, update status, and recent requests. It refreshes in real time over server-sent events.

What to watch

• Traffic rate - requests per second and tokens per second over the last minute.
• Latency and errors - p50 latency and error rate, each with a small sparkline.
• Telemetry rail - inference backend health, GPU snapshot, and update status.
• Residency timeline - recent model load/unload activity and peer residency.
• Running models - local models currently resident plus configured peers.
• Recent requests - the latest proxied exchanges, linked to the request detail view.

How it updates

The dashboard reads initial state through the admin API, then receives live invalidation events over server-sent events. Request rows, model state, GPU snapshots, and update checks can refresh without a full page reload.

Related pages

• Models for lifecycle controls and per-model detail.
• Requests for searchable request history.
• System for runtime, database, proxy, and GPU health.